Özgür Izmirli

Revitalizing computing education through free and open source software for humanity

The humanitarian focus of socially useful projects promises to motivate community-minded undergrads in and out of CS.

Haptic and sound correlations: pitch, loudness and texture

People process information through multiple modalities and many studies have been done that show interaction among the auditory, haptic and visual modes. This paper reports on a correlation between the haptic property of roughness and each of the audio properties of pitch and loudness. There were four experiments: given a roughness, match a loudness level to it and vice versa and given a roughness, match a pitch to it and vice versa. Subjects showed a strong correlation in the first two experiments, associating louder sounds with rougher textures and softer sounds with smoother textures. They also showed, although not always in the same direction, a strong correlation between pitch (high/low) and roughness. Applications of this study are to multimodal user interfaces and to music cognition.

A Road Sign Detection and Recognition System for Mobile Devices

We present an automatic road sign detection and recognition service system for mobile devices. The system is based on a client-server architecture which allows mobile users to take pictures of road signs and request detection and recognition service from a centralized server for processing. The preprocessing, detection and recognition take place at the server end and consequently, the result is sent back to the mobile device. For road sign detection, we use particular color features calculated from the input image. Recognition is implemented using a neural network based on normalized color histogram features. We report on the effects of various parameters on recognition accuracy. Our results demonstrate that the system can provide an efficient framework for locale-dependent road sign recognition with multilingual support.

Estimating the Tonalness of Transpositional Type Pitch-Class Sets Using Learned Tonal Key Spaces

This paper proposes a method to estimate the tonalness of a pitchclass set using transpositional types. For each set under consideration the method uses the corresponding transpositional type to generate note collections from acoustical instrument sounds and subsequently calculates a learned projection of that set into a low-dimensional space. The structure in this representation is then compared to the structure of a low-dimensional tonal key space learned from audio recordings of labeled tonal music. The term tonalness refers to how strongly the input suggests congruence to pitch use distributions in common-practice tonality. The method is tested on pitch-class sets of cardinality 3 and compared with measures from other work.

Constructing Geometrical Spaces from Acoustical Representations

This paper presents several models for constructing geometrical spaces from acoustical representations. Through specific spectral representations and associated distance measures each model is designed to highlight or ignore certain types of relationships within the given pitch sets. Dimensionality reduction is employed to obtain low dimensional embeddings from spectral representations. The viability of these models is demonstrated for the resulting low dimensional embeddings with respect to a number of group actions including octave shifts, permutation, transposition and inversion.

Elephant censusing via geophone arrays: A visual approach for linear arrays

The problem of censusing elephants via geo-phone recordings of their footsteps along a linear array positioned along a path leading to a watering hole is considered. We propose an intuitive and graphically motivated technique that delineates passings of herds of elephants. Additionally, a measure of dispersion of the localization estimates is proposed that correlates to the size of the passing herd. We illustrate the methods using both real data and a simulation study.

Choosing a Charging Station Using Sound in Colony Robotics

This research is part of a study that is investigating methods for supplying power to autonomous robots in remote locations. One possibility is to provide autonomous charging stations fueled through solar power. This paper reports preliminary results for a system being proposed to help a robot determine which station, among those available, would be its best source of power. The system uses distinct sound frequency bands to determine the available stations and uses the characteristics of sound and the environment as a type of analog computer to help compute, considering each stations distance and available energy, the best station to track towards.

Attribute correlations between haptic and auditory modalities

Correlations between attributes in the haptic domain and those in the auditory will assist in multimodal applications. This paper looks at the auditory attributes of pitch and reverberation and the haptic attributes of vibration, size and viscosity and seeks to correlate changes in one of the modalities with changes in the other. A total of 4 pairs of experiments were performed. The results indicate significant correlations in 3 of the pairs: pitch/vibration, pitch/viscosity, reverberation/size. There was no significant correlation found between reverberation and viscosity. These results have significance for designers of multimodal interfaces in virtual environments.

Signatures and pattern anchoring for score following

Abstract The process of score following requires that the automated follower be capable of making robust decisions during a performance, especially in light of the number of errors that need to be accounted for between the live performer and the follower. Based on the concept of variable-length signatures for monophonic symbolic input, we define a causal method for tracking score location. The signature lengths are calculated prior to performance based on their novelty in a predetermined locality within the score. The score following model is event based, but it also uses an inertial mechanism to pace itself in the presence of contradictory information. Results are derived from the output of the score follower using a graphical display that depicts the associations between performed notes and the score. This display aids in visual evaluation of the performance of the score following process and, specifically, its capability of recovery following any of the performance aberrations.